Monkey run experiment, revenue goes up, monkey get promotion.

At this point I cannot be alone in thinking that what we are seeing is a runaway AI system?

Let me just speculate wildly here:

Seems no-ones Google is the same and how do one manage approximately 4 billion [1] a/b/c/etc experiments if not using artificial intelligence and reinforcement?

Now imagine 4 billion agents ("monkeys") who flip bits everywhere and every time one of them is unusually successful that pattern is reinforced across a part of the "population" of agents.

Every time one of them does something particularly stupid Google relies on users to report an error and swaps out that agent. (This explains why if you report an error in search it appears to magically be fixed 15 minutes later.)

Let's say this has been running since way before 2008 and 2008 just happened to be the time it got completely out of hand:

Mostly it works very well economically speaking but search quality suffers year by year and no one can do anything about it since by now so many optimizations have been applied that for a decade no single human has been able to really understand it.

The AI - using end to end optimization techniques - learns to prioritize ad-heavy sites (since thise are more likely to result in positive reinforcement in the form of juicy ad impressions or clicks) and sites that doesn't really answer the question (since that brings the user back to Google for a new chance to serve ads both in search results and on other high ranking pages.)

In the same manner the system learns to avoid easily understood clean pages without ads as that means game over for this query.

By now everyone who knows anything about it sees that it broken but nobody dares to fix it since that would result in a massive revenue loss.

[1]: assuming half the world uses Google